In order to improve the fault diagnosis ability of the system and its accuracy, with previous experience in this article is based on data Rolling health management, this paper presents a novel based on multiple hidden Markov models and Kazuo artificial neural network algorithms and methods of combining ant colony to be used to diagnose and detect bearing faults, which uses HMM and pattern recognition method by combining the bearing vibration signal feature extraction, in the frequency domain analysis of the aging phenomenon, namely the historical data and the new data fault diagnosis and testing, while HMM and ANFIS fault prediction is to estimate the remaining useful life and the life. The experimental results show that the method of HMM and pattern recognition can be used to diagnose and predict the faults. The method can reduce the computational complexity and improve the accuracy of diagnosis, through the different fault diagnosis example elaborates on HMM-based fault diagnosis method effectiveness and feasibility.